THE STRENGTH OF STRONG TIES
A MODEL OF CONTACT-MAKING IN POLICY NETWORKS
WITH EVIDENCE FROM U.S. HEALTH POLITICS
Daniel Carpenter, Kevin Esterling and David Lazer
ABSTRACT
Interest groups establish contacts with each other as a way of gaining useful policy information, and in this article we develop and test
a model to explain this political phenomenon. Our simulation
model suggests that when few need information, groups will
pursue an acquaintance strategy by investing time and resources in
gaining `weak tie' political acquaintances rather than in gaining
`strong tie' political friends, but that as the collective demand for
information rises, groups increasingly follow a chum strategy,
placing greater emphasis on establishing strong ties. We test these
hypotheses in an analysis of inter-organizational contact-making
in U.S. health politics, using the data of Laumann and Knoke
(1987), with OLS regressions of average group contacts over lobbying events over time and maximum likelihood count models of
contacts across interest groups. Both analyses show that as collective demand for information increases, interest groups place greater
priority on establishing strong ties, even while controlling for
organizational attributes such as budget, mobilization capacity
and organization age. The results suggest some conditions where
policy networks in the aggregate are less likely to distribute information ef®ciently, and, in particular, that policy networks are less
ef®cient at distributing information when information is most in
demand.
KEY WORDS . acquaintance . information . interest groups .
weak ties
`More than mere technical experts, network people are policy activists who know
each other through the issues. Those who emerge to positions of wider leadership . . . are experts in using experts, victuallers of knowledge in a world hungry
for right decisions.' ± Hugh Heclo (1978: 103)
Rationality and Society Copyright & 2003 Sage Publications (London, Thousand Oaks,
CA and New Delhi), Vol. 15(4): 411±440. [1043±4631(200311)15:4; 411±440; 036620]
412
RATIONALITY AND SOCIETY 15(4)
More than two decades ago, Hugh Heclo (1978) called the attention
of political science scholars to the importance of `issue networks' in
any explanation of the development of modern public policy. Peak
associations and `iron triangles' alone could not explain the dynamic
of U.S. policy development. Heclo (1978: 104) writes:
Increasingly, it is through networks of people who regard each other as knowledgeable, or at least as needing to be answered, that public policy issues tend to
be re®ned, evidence debated, and alternative options worked out ± though
rarely in any controlled, well-organized way.
In subsequent decades, however, political science has developed
surprisingly little theory to explain how it is these networks evolve
among interest groups and other political organizations. Networks
are not designed, but are the concatenation of myriad rational decisions as to whether to talk or not to talk, to acquaint or not to
acquaint (e.g. Knoke 1990; Knoke et al. 1996; Verbrugge 1977;
Padgett and Ansell 1993). Without a theory of how these individual
decisions are made, one may conclude with Heclo that contemporary lobbying politics is disorderly and unpredictable, creating
problems for democratic governance.
The interest group research has shown that congruence of policy
interests is an important determinant of who talks with whom in
policy networks (Hojnacki and Kimball 1998; Koenig and Braunninger 1998). Here we examine how organized groups allocate their
time in establishing different types of contacts (which we label strong
and weak ties) in a network. Allocation of time among contacts is an
important facet of the lobbyist's job, determining in part how
informed he or she is on current policy issues.1 Lobbyists ®nd this
access to policy information critical; in particular, they may get
`access' to policy-makers by virtue of their demonstrated ability to
convey credible and useful information to Congress or the bureaucracy (Milbrath 1963; Bauer et al. 1972; Heclo 1978; Hansen
1991). Given limited time, lobbyists face a trade-off between developing and maintaining trusted informational contacts (`strong ties')
and more distant acquaintances (`weak ties'). To optimize the tradeoff between strong and weak ties, we argue that interest groups must
take into account a general rule of political and social life: friends
help their friends ®rst, their acquaintances later (see Uzzi 1996).
We demonstrate in this article that as the collective demand for
information among interest groups increases, lobbyists will invest
more time in making strong ties than in making weak ties.
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
413
In earlier work, Granovetter (1977) has shown that weak ties distribute information more ef®ciently than strong ties, since strong ties
tend empirically to be intra-clique and so less likely to provide new
information. Our analysis offers an amendment to Granovetter's
`strength of weak ties' argument: in competitive informational
environments, which often characterize national level health politics,
information tends to be concentrated in the relatively inef®cient
strong tie network. Our theory and analysis therefore suggest conditions for the relative ef®ciency or inef®ciency of the distribution
of information in an issue network. In this argument, one important
aspect of Heclo's problem of issue networks for national policy
development is conditional: in some identi®able circumstances,
policy information is likely to be undersupplied and so debates on
important policy topics are likely to be relatively under-informed.2
To formally establish these conditions, we develop a simulation
model and test the implications of the model with data on the communications network connecting national-level health policy lobbyists in the 1980s. These data were collected by Laumann and Knoke
in their study The Organizational State (1987). Using these data, we
test our theory in two ways. Our ®rst test examines the evolution of a
single policy network over time, covering 85 lobbying events in
health politics in the 1970s. We ®nd that the greater the interest
among groups in the lobbying event, the greater the average strongtie investments of groups involved in that event. Our second test
compares lobbying organizations. Using maximum likelihood count
regressions, we ®nd that groups involved in events that many
other groups are also interested in (i.e. events with greater demand
for information) tend to invest in more strong ties.
We develop these ®ndings in ®ve sections. We ®rst outline the
assumptions underlying our approach to the study of policy networks. In Section II we develop a simulation model of contactmaking based on these principles and derive model implications.
We offer hypotheses from these implications in Section III and discuss the data we use to test them. In Section IV we review the results
of our statistical analyses of contact-making in the health policy
domain. In the conclusion we discuss the broader implications of
these ®ndings for the study of policy networks.
414
RATIONALITY AND SOCIETY 15(4)
I. Lobbying and Networks
Interest groups seek to maximize their chances of being informed on
the issues in which they are interested (see, e.g., Milbrath 1963;
Bauer et al. 1972; Hansen 1991; Austen-Smith 1992; Austen-Smith
and Wright 1992, 1994; Rasmussen 1993). Being informed, on both
substantive and tactical issues, increases interest groups' ability to
manage and control the policy process. We develop a micro-level
model of how interest groups develop contacts in order to maximize
the amount of information they obtain through these contacts. We
build on Boorman's (1975) ground-breaking work on competitive
information search in an employment context. Like Boorman, we
assume that much useful policy information is socially distributed
in the lobbying community, and that groups acquire information
largely through contacts with other groups (Heclo 1978).3 Often
the maintenance of contacts with other groups can yield more
useful information and at lower cost than conducting research
anew for each policy issue. As Bauer et al. state, interest groups
are `nodes in the communications process', and as a general consequence, `what they knew or failed to learn, what they heard or did
not hear, what they said or failed to say, had a profound effect on
what other people learned, heard, or said' (1972: 325).4
How do interest groups go about choosing contacts? Granovetter's `strength of weak ties' hypothesis (1973) suggests that lobbyists are better off investing time in acquaintances (or `weak ties')
because they are more likely to hear new or novel policy information
through weak ties than through strong ties. This hypothesis follows
from an empirically observed feature of networks: strong ties tend to
be densely socially interconnected, and so the information that ¯ows
through strong ties tends to be redundant. In contrast, weak ties
tend to act as bridges between tightly knit cliques, and so it is
through `weak ties' that novel information diffuses most rapidly
across a network (see also Schneider et al. 1997: 1205 and Browne
1988: 55).
The extent to which groups gain information through social contacts, however, depends critically on the social rules and incentives
for information-sharing in policy networks. We argue that, in
competitive information environments, there is a tendency for friends
to share information with their close and trusted friends before
they share information with acquaintances. This social rule is well
illustrated by a lobbyist interviewed in the classic study of Milbrath:
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
415
My contacts trust me, and I think their trust is well placed. Most of the things they
tell me are not of a secret nature; it's just a development that they have discovered
which they think I would be interested in. It is very dif®cult to get information if
you go out digging for it. . . . Actually, you get much better information from
people who know you, know what your interests are, and know that they can
trust you. (1963: 260)
And as Granovetter notes (1982: 113): `Weak ties provide people
with access to information and resources beyond those available
in their own social circles; but strong ties have greater motivation
to be of assistance and are typically more easily available.'
There are three reasons why strong ties might receive preferential
treatment in the context of policy networks. First, there is a `bandwidth constraint': there are limits to how many others one can
communicate with in a day.5 That is, typically one cannot simply
broadcast a bit of information to all of one's contacts who might
bene®t from having that information instantly. Instead, one will
communicate that information over a period of time to one's contacts. Given this constraint, each actor will need to prioritize who
receives information ®rst.
Second, even if a source had unlimited capacity to convey information, the source would be more likely to know that a given
strong tie was interested in a piece of information than a given
weak tie, as the Milbrath quote above suggests.
Third, if possession of asymmetric information confers a distributional or political advantage in politics (e.g. Austen-Smith and
Wright 1992 and Hansen 1991), then a source would likely give
that information to strong ties before giving it to a weak tie or
acquaintance.
A concrete illustration from recent national level health policy
helps to establish the argument.6 The expected number of managed
care plans that will leave the Medicare program in a given year, and
hence the number of elderly who are likely to lose coverage in their
chosen plan, is based on the content of different reform proposals
before Congress. These proposals often involve relatively arcane
topics, such as risk adjustment methodology, capitated payment
rates, program user fees, and mandated prescription drug coverage.
Assume that the American Association of Health Plans (AAHP), the
leading trade association for managed care plans and health maintenance organizations (HMOs), knows the number of plans that will
exit the Medicare program based on the different proposals before
Congress. The AAHP faces high costs of communicating policy
416
RATIONALITY AND SOCIETY 15(4)
Figure 1. Illustration of strong tie priority rule
information on these arcane topics, since the various political and
policy rami®cations of various proposals might require an afternoon
session to explain. Furthermore, assume the AAHP has a strong tie
to the Health Insurance Association of America (HIAA) and a weak
tie to AARP (formerly known as the American Association for
Retired Persons). In this case, the AAHP is more likely to know
that HIAA wants to know about managed care exit, is more inclined
to invest the time and resources to explain the topic to HIAA, and is
more inclined to confer the potential bene®ts for knowing this information to HIAA, when compared to the AARP. That is, the AAHP
will likely send information to HIAA before it sends information to
the AARP.
The `strength of weak ties' hypothesis and the `strong tie priority'
rule together imply that there is a trade-off between seeking to gain
better or more novel information through establishing weak ties,
and seeking to increase the priority one's contacts place in sending
you information. It is also reasonable to assume that a trusted
friendship requires more time to form and maintain than an
acquaintance; a lobbyist can make more weak than strong ties,
but the added expense of strong ties brings preferential treatment
(e.g. see Browne 1988: 55). It is this trade-off between strong and
weak ties that we formally model and statistically analyze in the
remainder of the article.
II. Simulation Model
Let a policy network consist of N interest groups with potentially
common interests (but not common positions) in a set of issues,
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
417
and let an event consist of the full debate and information transfer
that occurs among these N groups when the government considers
action in a relevant policy domain. For example, the policy
domain might be health, the issue might be healthcare ®nancing,
and an event might be the 1994 Clinton healthcare plan debate.
We assume that for a given piece of information, in a population
of N interest groups, there is probability that any given interest
group has the information at the beginning of an event. Our
model describes the process by which this information then diffuses
through the informational network that connects interest groups.
If an interest group is not one of the initial groups with information,
if and when it will acquire that information depends on its connections to other groups. The trick for an interest group, then, is to position itself in the network so that information will ¯ow to it as quickly
as possible.7
The `collective demand for information' among lobbyists is the
fraction of the network population that is interested in the information. In the lobbying context, there is clearly great variation by event
in how many groups are interested in acquiring information. Some
issues are more complex and general than others, and this will determine the general demand for information on that issue. More health
policy groups are likely to be interested, for example, in hospital cost
containment than in DNA research. We therefore assume:
Assumption 1: The collective demand for policy information varies
across events.
In the simulation model, we assume that the fraction of the population of interest groups that needs information, , varies over events
and may equal 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, or 1.8
We assume that each event is made up of R rounds. Given the
costs of and other constraints on communicating information as
described above, assume that during each round a group with information may pass that information selectively on to one other group.
For any pair of groups, A and B, there may be no tie from A to B, a
weak tie from A to B, or a strong tie from A to B. Given the strong
tie priority rule set out above, we assume that:
Assumption 2: If A has information, it will ®rst send that information
to a strong tie, and only if no strong ties need information will it
send it on to a weak tie.
418
RATIONALITY AND SOCIETY 15(4)
In the simulation, we add the assumption that all groups have a time
constraint, T, on how many ties they can form (otherwise groups
would form strong ties to all other groups), such that:
T ˆ W ‡ S, where > 1 (since strong ties are more expensive
than weak ties).9
Lobbyists may therefore invest in a continuum of potential strategies. At one end of the continuum, lobbyists might invest their
time in gaining many `weak tie' acquaintances in a loose network
of contacts who know something about the policies in which they
are interested. We label this an acquaintance strategy. At the other
end of the spectrum, lobbyists may invest their time in gaining a
few valued `strong tie' lobbying partners, political friends in whom
they will invest more time and cultivate more trust, and from
whom they expect to receive more information in return. We call
this the chum strategy. We seek to model the optimal trade-off in
the mixture (or the choice not to mix) between these two pure
strategies.
Finally, we assume that over time interest groups learn effective
strategies for obtaining information:
Assumption 3: Interest groups emulate the relatively successful networking strategies of other interest groups in the system.
This emulation process is modeled with a genetic algorithm (e.g. see
Goldberg 1989; Holland and Miller 1991; Andreoni and Miller 1995;
Axelrod 1997; see generally Kollman et al. 1995, 1998). The basic
assumptions of the genetic algorithm are that after a ®xed number
of events the relatively successful strategies are spliced together to
form new strategies. After this process is iterated a number of times,
the population of strategies will tend to converge on a single,
highly successful strategy (see Appendix 1 for a summary of the
genetic algorithm as utilized in this article).
Figure 2 summarizes how we implement these assumptions in the
simulation model.10
The key question we are asking is how the allocation of time
between strong and weak ties is affected by the demand for information () in a given event. Below we summarize a set of 1000 simulations, with 100 interest groups, where strong ties `cost' ®ve times
as much as weak ties (i.e. ˆ 5), groups can have up to 10 strong
ties (or 50 weak ties), there are 5 rounds per event, 40,000 events,
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
419
å Basic model parameters. There are N groups. Each group is
informed at the beginning of an event with probability . Each
group needs information at the beginning of the round with probability . varies from event to event, and may equal 0.1, 0.2, 0.3,
0.4, 0.5, 0.6, 0.7, 0.8, 0.9, or 1. is publicly known, and is chosen
randomly at the beginning of an event. Each group has a time
budget, T, for establishing contacts: T ˆ W ‡ S, where > 1.
å Tie investment decision. At the beginning of an event, a group
chooses how many strong and weak ties to establish, based on .
These ties are then randomly chosen at the beginning of the event.
å Information dissemination. There are R rounds in an event. During
each round, groups that have information may pass that information on to one other group that needs information. In each round,
if a group has information, it will ®rst check among the strong-tie
groups to whom it sends information. If one or more of those
groups need information and one or more weak ties needs information, it will send information to one weak tie.
å Adaptation. In the ®rst event of the simulation, strategies are
randomly determined. In subsequent events, more successful strategies are emulated (see Appendix 1 for description of genetic
algorithm used to adapt strategies).
Figure 2. Key simulation assumptions
after every 200 of which adaptation occurs via the genetic algorithm.
In each event there are 5 rounds during which information spreads.
The performance of each group (used in the genetic algorithm) is
measured by the proportion of the rounds in which the group was
informed when it needed information. Thus, if a group needed information in event 5, and received information in round 3 of event 5,
it would receive a score of 0.6.
In these simulations each group has what we call a `network strategy string' (NSS) ± a list of 10 integers between 0 and 10. The ®rst
digit in the list indicates how many strong ties the group will
choose if ˆ 0:1, the second digit how many strong ties the group
will choose if ˆ 0:2, the third digit how many strong ties the
group will choose if ˆ 0:3, and so on, until the tenth digit,
which indicates how many strong ties the group will choose if
ˆ 1 (we assume that each group devotes the remainder of its
420
RATIONALITY AND SOCIETY 15(4)
time to weak ties).11 The population of NSSs evolves through the
run of a simulation, converging on NSSs that maximize the amount
of information each group receives. At the beginning of each simulation, the values of the NSSs are randomly chosen from a uniform
distribution. Thus, at the beginning of these simulations, the average
group selects ®ve strong ties (and 25 weak ties) for each value of .
As we mention above, the strength of weak ties literature asserts
that strong ties tend to be more intra-clique than weak ties. In the
interest group universe, one might imagine that strong ties are
within coalitions, and weak ties between coalitions. A key issue,
then, is whether this empirical regularity affects the model results.
In the analyses below, we present the results model results under
two conditions: 1) where strong ties are distributed randomly, and
2) where the 100 interest groups are partitioned into 5 cliques of
20 groups each; strong ties are still randomly allocated, but just
intra-clique. Weak ties, in both cases, do not depend on the partitioning of cliques.
The question now is: What is the relationship between and the
number of strong (and weak) ties each group selects at the end of
these simulations? Figure 3 summarizes the average number of
strong ties selected at the end of these 1000 simulations under both
conditions for the partitioning of strong ties. First consider the case
where strong ties are not partitioned. Figure 3 demonstrates a strong
pattern ± the higher the values of the more strong ties a group
selects. For ˆ 0:1, the percentage of the time budget of groups
invested in strong ties, averaged across all groups in the 1000 simulations, is 38%, where this number increases monotonically as approaches 1. At ˆ 1, almost all the time budget invested by a
group in ties to other groups is in strong ties (95%).
This ®nding mirrors Boorman's (1975) argument that strong ties
are more valuable in the employment search context when unemployment is high. There is a solid intuition here: as more
groups need information, the probability that a group with information will pass it on to a weak tie drops precipitously. Consider
a group from whom 5 other strong-tie groups and 25 weak-tie
groups would like to get information. If ˆ 0:1, the probability
that that group would pass information on to any weak-tie group
at the beginning of an event is 59%. If ˆ 0:5, this probability
drops to 3%, and for ˆ 0:9, to 0.001%! The 5 strong ties present
a barrier that grows exponentially with . Our ®rst result is
therefore:
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
421
Figure 3.
Result 1: The more interest groups seeking information regarding an
event, the more strong ties (and fewer weak ties) those groups will
seek.
The Boorman study imposes the assumption that there is no systematic difference between the structure of the strong-tie and weaktie network. We are interested in the optimal trade-off between
strong and weak ties assuming that strong ties tend to be intraclique and so are less likely to disseminate new information.
Figure 3 shows that when strong ties are partitioned into cliques,
groups do indeed systematically invest in fewer strong ties than in
the condition where strong ties are not partitioned into cliques.
This ®nding replicates Granovetter's key insight: the intra-clique
nature of strong ties reduces their value as sources of information.
But even in this extreme scenario, where strong ties are exclusively
intra-clique, the investments a lobbyist will make in strong ties
increase with the level of information demand in a given event. The
intra-clique bias does have a disproportionate effect on the value
of strong ties for high values of , so the relationship between and number of strong ties levels off for very high levels of (for
values 0.5 or greater). This is likely because as the density of intraclique ties reaches this threshold, information diffuses so quickly
within the clique that an additional tie produces very little bene®t.
Our second result is therefore:
422
RATIONALITY AND SOCIETY 15(4)
Figure 4. low information demand
Result 2: The tendency of strong ties to be intra-clique attenuates but
does not reverse the importance of strong-tie contacts in a high
information demand environment.
Both results 1 and 2 imply the same testable proposition: the more
groups collectively demand information, the greater the advantage
to having friends rather than acquaintances as contacts.
Thus, returning to the example from Figure 1. The relative value
of the strong tie AAHP has to HIAA as compared to the weak tie
AAHP has to AARP is contingent on the demand for information
of the other actors AAHP is connected to. In Figure 4, only one
other actor (a weak tie) that AAHP is connected to needs information that the AAHP has (signi®ed by the shading). The AARP would
be ®rst or second in line to receive information that the AAHP has.
In Figure 5, information demand is far higher (with 8 rather than
3 actors needing information). The value of HIAA's strong tie will
drop (it is guaranteed being just one of the ®rst four actors to receive
information); as will the value of AARP's tie (it will be somewhere
between the ®fth and eighth actors to receive information). In a
world where at critical junctures events move quickly, those at the
end of the queue may never receive the information in time to use it.
III. Hypotheses and Data for the Empirical Analyses
The above model offers a simple and testable implication: the
greater the collective demand for information among groups,
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
423
Figure 5. High information demand
measured as the relative number of groups in the population seeking
a particular piece of information, the greater the tendency of groups
to invest in strong ties relative to weak ties. We test this implication
in two ways, with organization-level data and with ecological eventlevel data.
Event-Level Hypothesis
Our data encompass interactions over many lobbying events, which
are occasions for lobbyists to approach the government as it is
considering an of®cial action on an issue (e.g. when a committee is
marking up a bill on hospital cost containment). We expect that at
the event level, the more groups that are interested in the lobbying
event, the greater the average number of strong ties of the groups
active during that event.12
Hypothesis 1A: The average number of strong ties to groups
involved in a given event is positively related to the number of
groups interested in that event (the level of `collective information
demand' in those events).
In the data that we use to test this relationship, the network data are
effectively aggregated over many events. That is, groups do not start
with new networks with each event, but can only tweak their networks over time depending on their informational needs. This
should attenuate (but not eliminate) the predicted relationship.
424
RATIONALITY AND SOCIETY 15(4)
Organization-Level Hypothesis
Hypothesis 1A implies that at the event level, groups should establish more strong ties if they are involved in a high information
demand event. A typical interest group is involved in many different
events, however, and at the organization level a group's strong-tie
investment should be a function of the overall pro®le of events in
which the group is involved. We therefore produce a second test
of result 1 of the model at the individual group level. If the demand
for information varies across events, and groups are differentially
involved in those events, then we should expect a positive correlation
across interest groups between the level of information demand in
the set of events they are involved in and their group-level strongtie investments. That is, groups that are disproportionately involved
in events in which there is a high demand for information should
have more strong ties than groups that are not.
Hypothesis 1B: The number of strong ties to a group is positively
related to the average level of information demand in the pro®le
of events in which that group is involved.13
Controls
The organization-level model controls for characteristics of the
organization that also may affect contact-making behavior, and
the event model controls for the aggregate characteristics of the
set of organizations involved in each event. Our simulation model
assumes that groups have a time budget in which the choice between
strong and weak ties is a zero-sum trade-off. As the time budget is
unobservable, there is no conceivable way to reconstruct it given
our data. Because groups with a large time budget will establish
more of both strong and weak ties, we enter the number of weak
ties in the model as a partial proxy for a group's time budget. If
we had a measure for the budget, and included that measure in
the equation below, we would expect a negative relationship between
the number of weak ties and the number of strong ties. In the
absence of a measure for this time budget, we expect the relationship
to be (spuriously) positive, since the time budget should be positively
related to both. By including weak ties as a control, the impact of
independent variables we include in the equation below should be
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
425
interpreted as affecting the balance of investments in strong versus
weak ties.14
In addition, we control for a number of group characteristics in
our data set. We have several measures of the political clout of an
interest group, on the assumption that groups with more political
in¯uence can gain access to the government irrespective of their relative credibility, as, for example, through campaign contributions
(Bennedsen and Feldmann 1998). Groups with greater clout may
therefore have the luxury to pursue weak ties, which confer little
information for a high information demand lobbying event, but
which confer more information on average as events come and go.
To control for this, we include measures of each group's lobbying
budget, and estimates of its capacity to mobilize both its members
and the public as issues arise (see Table 1). We control for the
group's age, assuming that older groups have had more time to
establish strong contacts. Finally, we control for whether or not
the group is an issue-oriented public interest group.
Data
We test the informational contact-making model in healthcare
politics, where, at the time encompassed by our data, technical
policy information was at a premium (Hyman 1971; Schwartz
1972; Alford 1975; Bauman 1976).15 We employ the rich data
collected by Laumann and Knoke (1987) for their book The Organizational State. Laumann and Knoke surveyed informants from an
exhaustive list of the most prominent and in¯uential health lobbying
organizations.16 Their sample of health lobbying organizations
includes industry associations (such as the American Insurance
Association and the Pharmaceutical Manufacturers' Association ±
now PhRMA), professional societies (American College of Cardiology, American Medical Association), health interest groups (Coalition for Health Funding, Arthritis Foundation), as well as more
general interest groups, business ®rms, and relevant government
agencies and congressional committees. These data document the
full network of communication ties from each of these health policy
organizations to each other, lobbying groups' internal organizational characteristics, and each organization's lobbying activity in
85 policy events that occurred between 1973 and 1980.
In Table 1 we summarize the dependent variables and how they
were constructed, and in Table 2 the independent variables.
426
RATIONALITY AND SOCIETY 15(4)
Table 1. Dependent Variables
Strong ties ± group level A count of the number of a group's strong ties through
which it receives information. A strong tie occurs if some
other group reports that it gives advice to the group as
part of a `trusted exchange of sensitive and con®dential
advice'.
Strong ties ± event level
The average number of group-level strong ties for the
groups involved in an event.
Table 2. Independent Variables
Theoretical variable: Collective demand
for information ±
event level
Collective demand
for information ±
group level
Control variables
Weak ties to other
health lobbying
groups
The fraction of all groups that reported a moderate or
high level of interest in the event, multiplying the fraction
by 100 to allow for percentage-point interpretation of the
coef®cients.
The average of the event-level information demand of the
events in which the group participates.
The number of a group's weak ties through which it
receives information. A weak tie occurs if some other
group `regularly and routinely discusses national health
policy matters' with the group but does not report
having a strong tie with the group.
Organization budget
Recorded in 1980 dollar amounts, logged to improve
maximum likelihood model convergence.
Estimated
mobilization capacity
The number of times other groups name the organization
as one whose in¯uence comes from its ability to `mobilize
its members or employees to support a proposal'.
Estimated public
mobilization capacity
The number of times other groups name the organization
as one whose in¯uence comes from its ability to `mobilize
general public opinion to support a proposal'.
Organization age
Recorded in years.
Public interest group
Coded 1 if the group describes its main function as a
public interest group voluntary organization, 0 otherwise.
Note: In the event-level (ecological) model, the control variables are measured as
the average of these organization-level control measures for all groups participating
in the event.
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
427
IV. Empirical Analysis: Contact-Making in U.S. Healthcare
Politics
We now examine the impact of information demand upon tiemaking in two ways. First, we examine our network over time, by
looking at a sequence of events in health policy. In a second analysis,
we decompose the network and examine strong ties at the group
level.
Results from Events-Level Analysis (Network `Time Series')
We ®rst assess the relationship between the information demand and
lobbyist contact-making by examining a set of 85 lobbying events
that occurred in the health policy domain in the 1970s. Table 3
summarizes the results of a regression of strong ties on information
demand with the full set of ecological controls, and with a test of
temporal autocorrelation (Model 4). Table 3 also summarizes
three alternative speci®cations: a reduced model of strong ties on
collective demand for information and weak ties (Model 1); the
reduced model with a test of temporal autocorrelation of errors
(Model 2); and a regression of strong ties on information demand
and the full set of control variables but without the test of temporal
autocorrelation (Model 3).17
The results summarized in Table 3 offer strong support for
Hypothesis 1A: the higher the level of information demand in an
event, the more strong ties those groups involved in that event have.
This relationship is signi®cant at p<0.001, where a 1 SD increase
(14.66 percentage points) in the fraction of groups interested in an
event is associated with nearly 2 additional strong ties for the average group involved (the average group had 6.33 strong ties). While
the signi®cance of the budget (p<0.05) parameters suggests that
models 3 and 4 include the preferred speci®cation, in the other
two speci®cations the relationship between information demand
and strong-tie investment is substantively similar and signi®cant at
p<0.01 or better.
Results from Group-Level Analysis
The analysis in this section takes the group as the unit of analysis.
Table 4 presents two negative binomial regressions.18 The ®rst
includes all of the group-level independent variables listed in
Model 1
‡ weak ties
5.4129*
(2.7202)
10.3966**
(2.5942)
0.5030**
(0.0520)
±
±
Variables
Constant
Aggregate demand for information ()
(Test of H1: Proportion of groups reporting
moderate or strong interest in the event)
Average weak ties of groups participating in
event
Average budget of groups participating in
event
Average age of groups participating in event
±
±
0.5122**
(0.0516)
11.0192**
(2.5978)
6.0579**
(2.7133)
Model 2
‡ weak ties
AR(1)
0.0036
(0.0324)
1.26e±07**
(3.77e±08)
0.4282**
(0.0672)
8.9339**
(2.5444)
4.4951
(3.7191)
Model 3
‡ all controls
0.0033
(0.0324)
1.26e±07**
(3.77e±08)
0.4289**
(0.0672)
8.9596**
(2.5457)
4.5181
(3.7182)
Model 4
‡ all controls,
AR(1)
Table 3. Relationship between information demand and strong-tie investments in the network: event-level analysis
(Prais-Winsten Regression Estimates)
428
RATIONALITY AND SOCIETY 15(4)
±
…"t †
(Autocorrelation parameter)
85 (82)
85 (82)
0.5448
51.27**
1.8521
Number of events (d.f.)
Adjusted R2
F
Durbin-Watson
1.9724
25.77**
0.5958
85 (79)
1.31
±
2.5513
(3.3442)
1.9804
25.82**
0.5963
85 (79)
1.31
0.0049
2.5390
(3.3411)
Dependent variable is the average number of strong ties among all groups participating in the event. (Standard errors in parentheses.)
Source: Laumann and Knoke (1987).
Note: ** Denotes signi®cance at p < 0:01. * Denotes signi®cance at p 0:05. (All tests are two-tailed.)
1.9852
54.15**
0.5586
1.62
0.0785
±
1.52
Marginal effects of [Additional strong ties ± for every group in the
network ± from a 1 SD increase in the aggregate
demand for information]
±
Proportion of public interest groups participating
in event
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
429
±
0.0057*
(0.0026)
±
±
0.0060**
(0.0014)
±
84
106.82** (4)
0.3174**
(0.0873)
84
65.04** (1)
0.0533
(0.0429)
±
0.0504+
(0.0265)
±
±
0.6077
(0.8333)
0.0143**
(0.0053)
3.7688+
(2.1114)
Negative binomial
0.5594
(0.4767)
0.0134**
(0.0022)
3.8575*
(1.5752)
Poisson
Reduced model
Source: Laumann and Knoke (1987).
Maximum likelihood count regressions (Poisson and negative binomial); dependent variable is number of strong ties of group i. (Standard
errors in parentheses.)
Note: ** Denotes signi®cance at p < 0:01. * Denotes signi®cance at p < 0:05. + Denotes signi®cance at p < 0:10. (All tests are two-tailed.)
N
2LLR
Dispersion
Public interest group
Estimated mobilization
Capacity
Estimated public
Mobilization capacity
Organization age (years)
64
112.05** (7)
0.0168
(0.0607)
0.0894
(0.0849)
0.0061
(0.0531)
0.0045
(0.0032)
0.0521
(0.2314)
0.2871**
(0.0887)
64
41.76** (1)
0.0072
(0.0327)
0.0985*
(0.0478)
0.0007
(0.0278)
0.0046**
(0.0017)
0.0723
(0.1191)
±
Average demand for information () across
issues in which group i is involved
(Hypothesis #1B)
ln(Budget)
Number of weak ties to groups
0.6395
(1.3838)
0.0193**
(0.0068)
6.1623*
(3.1002)
0.4891
(0.6349)
0.0177**
(0.0026)
6.5975**
(1.8175)
Constant
Negative binomial
Poisson
Variables
Full model
Table 4. The demand for information and lobbyists' `social' investment in strong ties
430
RATIONALITY AND SOCIETY 15(4)
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
431
Table 2. The second is a reduced model that excludes the organizational attribute variables, except for the group's budget, from the
®rst regression. For comparison, we also include the equivalent
Poisson regressions (the negative binomial models are to be preferred, since in both versions of the model the level of dispersion
of the dependent variable is signi®cant at p<0.001).
The group-level estimations provide support for Hypothesis 1B:
the greater the involvement of an interest group in issues in which
the information demand is high, the greater that group's investment
in strong ties, controlling for general propensity to establish ties
(measured as the number of weak ties). In both the Poisson and negative binomial regressions in Table 4, the impact of information
demand upon strong ties is estimated to be positive and statistically
distinguishable from zero. The impact of information demand is substantively signi®cant as well: in the full negative binomial regression
of Table 4, the marginal impact of a standard deviation increase in
information demand (0.04), with all other variables set to their
means, is 1.6 more strong ties (the marginal effect of a 1 percentagepoint increase is 0.39 ties). In other words, a 5 percentage-point
increase in the aggregate demand for information is suf®cient to
lead the average group to invest in two additional strong ties. This
effect is diminished in the reduced Model 4. In Model 4 a standarddeviation increase in information demand is associated with one
additional strong tie (the marginal effect of a 1 percentage-point
increase is 0.25 ties).
Another way of understanding the marginal effects of the aggregate demand for information is to assess the effects of a shift from
the minimum (0.05, or 5% of groups) to the maximum (0.28) of
the information demand variable. Holding all other variables at
their means, this shift would be associated with nine additional
strong ties for the average group. A shift from the ®rst quartile of
information demand (0.11, or 11% of groups) to the third quartile
(0.22) yields an additional four strong ties for the average group.
Two other interpretations of these results merit discussion here.
First, the marginal effects of information demand are consistent
across estimations. Notice that the marginal effects from the
group-level analysis of Table 4 are roughly comparable to the effects
observed in the event-level analysis in Table 3. While this is not
surprising, given that the two estimations are different cuts from
the same sample, it is nevertheless added evidence for the consistency
432
RATIONALITY AND SOCIETY 15(4)
of our results across two different analyses of the data. Second,
the observed level of information demand is precisely at the level
where our model suggests that the marginal effects of this variable
should be greatest ± when is low (see Figure 3). Notice that
approximately one in four groups seeks information about any
given policy event. If our measurements of are accurate, this is precisely the range from Figure 3 where increases in have the greatest
effect upon strong-tie investments.
Other Variables
As expected, the relationship between number of weak ties and
strong ties is consistently estimated to be positive and signi®cant.
This is likely a spurious relationship, where both are positively
related to the time budget that the organization devotes to making
ties. The estimated effects from the political clout variables are
statistically weak, but where signi®cant, the effects are in the
expected direction. For example, the ability to mobilize group membership is a measure of political clout, and in the organization-level
regression (full model, Poisson) this measure of clout appears to
reduce the need to establish strong ties. The effect of group budget
is positive and signi®cant in event-level analysis (and nearly so in
the reduced organization-level Poisson model), suggesting that a
large budget results in a capacity to develop the costlier strong ties.
The impact of organizational age on number of strong ties is estimated to be positive and signi®cant at p<0.05 in the reduced model.
The relationship is also substantively signi®cant: Taking marginal
effects estimates from the reduced negative binomial model, every
additional year of interest group age is associated with 0.037 strong
ties. The mean of group age in our sample is 40.37 years, with a standard deviation of 37.1 years. Hence a 1 SD boost in interest group
age is associated with an increase of approximately 1.5 strong ties
(1.37). Moving from the minimum of the age variable (4 years) to
the maximum (136 years) would yield an additional four strong
ties. This may be because strong ties take extended periods to
develop. Lobbyists can establish weak ties quickly ± over a `cocktail', perhaps.
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
433
V. Conclusion
As Milbrath's (1963: 260) respondent whom we quote in Section I
notes, `you get much better information from people who know
you, know what your interests are, and know that they can trust
you'. This quote usefully suggests that 1) interest groups gain
much of their political and policy information through their social
contacts, and 2) that the dissemination of information through
social contacts follows general social rules. In particular, groups
tend to give priority to their strong ties over their weak ties when
they choose to convey new information. Given this rule, we develop
a simulation model of information transfer among interest groups
and ®nd that the more demand for information about an event,
the more a lobbyist will follow a `chum strategy', forming more
strong ties and fewer weak ties. Our statistical analyses of lobbying
networks in healthcare politics in the United States provide empirical support for these arguments. We ®nd empirically that groups
involved in issues where the information demand is high tend to
invest more time in developing strong-tie `chums' than in weak-tie
acquaintances.
The model has interesting implications for the ef®ciency of the
network as a whole at informing its members. The network structure
that would result in the fastest diffusion of information is one in
which every group invested exclusively in weak ties; strong ties
tend to be intra-clique and weak ties tend to be inter-clique, weak
ties are more likely to transmit novel information. But because
groups are more likely to communicate information to their
strong-tie contacts, interest groups often invest in strong ties. As a
result, the network structure diverges from the `ideal', where the
greater the aggregate interest and involvement of interest groups
in an event, the greater the divergence. This prediction of a `network
failure', similar in structure to a prisoner's dilemma, raises empirical
and normative questions of how these individual contact strategies
affect the overall informedness of public policy. In particular, the
model suggests that the greater the need for a well-reasoned and
broad-based decision, the greater the tendency for a policy community to shatter into competing cliques that do not share information.
This article illustrates how the value of a particular strong or weak
tie is contingent on the other ties in the system, and on the ¯ow
of information along those other ties. In contrast to an ef®cient
434
RATIONALITY AND SOCIETY 15(4)
market for information, where all exchanges are arm's length, the
actual exchange of information among interest groups is governed
by social rules. We focus on strong-tie priority, but there are a
variety of other social rules endemic in social relationships that
deserve study and that are described in the literatures on social networks, including social exchange (Emerson and Cook 1978), homophily (e.g. for interest group applications, see Hojnacki and Kimball
1998; Koenig and Braunninger 1998), and transitivity (e.g. see
Holland and Leinhardt 1971, 1981). The common thread through
these literatures is the principle of interdependent interconnectedness ± that the interactions of any particular actor have rami®cations
that ripple through the system. It is this principle that we seek to
insert into the research on how interest groups seek and receive
information.
NOTES
We gratefully acknowledge helpful comments from John Brehm, Ken Kollman, and
Paul Teske. We retain exclusive rights, of course, to any errors contained herein.
1. In a growing literature, scholars have examined the strategic problems of credibility in games of information transmission (Austen-Smith 1992; Austen-Smith and
Wright 1992, 1994; Rasmussen 1993). In this article, we examine one of the preconditions of informational credibility: the problem of information acquisition.
2. The model we present here is based solely on the informational advantages that
networks confer. See Krackhardt (1992) for a more general discussion of strong
ties and weak ties.
3. This is in contrast to signaling models that assume that groups acquire policy
information at some set price. See for instance Austen-Smith and Wright's de®nition of an `information acquisition strategy' as a probability of payment for
information (1992: 34).
4. Kerwin (1994: 201) notes the informational advantage of contacts as well, ®nding that 100% of the interest groups he surveyed reported that they monitored
rule-making in part through communication with `colleagues in other groups'.
For 41.7% of these groups such communications occurred more than once a
week, and for 93.2% at least once a month. Similarly, Golden (1998: 258)
found that most of the groups she surveyed relied on issue networks to gather
information on rule-making.
5. These limits, of course, vary by the type of information and the technology available to the communicator. Thus, many interest groups place a fair amount of
information on the web (e.g. in the health policy area see the AMA website±
www.ama-assn.org). However, much of the information of relevance in the
lobbying world is unlikely to be disseminated through the web.
6. See, Mary Agnes Carey, `Medicare ‡ Choice Plan in Trouble as Insurers Line Up
to Leave', Congressional Quarterly Weekly Report, 20 May, 2000, p. 1183.
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
435
7. This model diverges from Boorman's (1975) model in two important ways. First,
Boorman makes highly restrictive assumptions about the structure of the network. The approach we take allows us to manipulate the structure of the network
to examine the impact of the structure on our ®ndings. Second, Boorman's model
assumes that information may only be effectively passed on to one actor. This
assumption makes more sense in the employment context, where only one
person may ®ll a job, than in the lobbying context, where consumption of information is often at least partially non-rival. The mathematical `apparatus' of
Boorman's model is therefore quite different from the model presented in this
article.
8. This corresponds to the parameter of the same label in Boorman's (1975) model,
which represented the unemployment rate. Note that if ˆ 0 no information
transmission occurs.
9. The assumptions regarding strong tie priority and the budget constraint, T, also
follow from Boorman (1975).
10. The code for the simulations is available from the authors upon request.
11. In this model, asymmetric ties are distributed randomly. For example, if A has
one strong tie, and it is to B, A gets information from B, and B does not necessarily get information from A. A more realistic assumption, but far more dif®cult
to implement, is that ties will typically be symmetric ± A and B exchange information ± and that tie formation and maintenance will, in part, be based on that
reciprocity.
12. Each interest group respondent was asked to indicate the nature of the group's
lobbying activity and inactivity on a series of governmental health policy decisions between 1973 and 1980. For example, the group may have actively lobbied
with a particular position, it may have just held a conference or seminar, or it may
have held a position but did not actively lobby.
13. Hypotheses 1A and 1B are logically equivalent, but in testing them both on our
data we are effectively checking the robustness of our empirical ®ndings.
14. By way of analogy, imagine one were modeling consumption of individuals,
where individuals' incomes varied, that no income was saved, and that individuals consumed only two goods: widgets and gadgets. Assume one was testing
whether some individual-level variable, x, affected consumption of widgets.
A regression of consumption of widgets on gadgets and x which yielded positive
coef®cients for both gadgets and x would simply indicate that, for a given level of
income, x resulted in increased consumption of widgets and decreased consumption of gadgets.
15. The relative importance of information in health lobbying can be inferred from
the `sample' of issues in the Laumann and Knoke (1987) data set, which includes
biomedical research issues such as DNA research and human experimentation,
the organization of healthcare delivery issues such as HMOs and hospital cost
containment, and regulation of food additives or medical devices.
16. The sample of organizations is not a probability sample drawn from a known
population, but rather an exhaustive list of organizations that are consequential
or highly visible in healthcare lobbying. An organization was deemed consequential if it appeared with some degree of regularity between 1973 and 1980 in newspaper articles covering healthcare politics, congressional hearings, amicus ®lings
in federal court, health-lobbying registration lists, or was named by a panel of
health politics experts. Laumann and Knoke used this method of non-random
436
17.
18.
19.
20.
RATIONALITY AND SOCIETY 15(4)
random selection since there is no known universe for sampling health-lobbying
organizations, and since it avoids selecting organizations on their degree of
connectedness (see Laumann and Knoke 1987: 95). The survey informant was
either the organization's executive director, government affairs or staff specialist.
Laumann and Knoke had an 89.4% response rate from health-lobbying organizations yielding 135 observations (see Laumann and Knoke 1987: 97±101).
An AR(1) speci®cation is added because the events analyzed in Table 3 are
ordered sequentially, although the inter-event times are not equivalent. One dif®culty with the dependent variable in Table 1 is that it is theoretically constrained
to lie above 0, whereas ordinary least squares makes no such assumptions. However, the mean of this variable (21.86 strong ties per group) is 4 SD (5.14) away
from zero.
The negative binomial regression estimator operates as follows. Let the dependent variable Si be the number of strong ties of group i. The observed Si are
assumed to be generated by a negative binomial distribution with mean and
variance …1 ‡ †. Letting gNB …Si † be the negative binomial probability distribution for Si we have gNB ˆ Pr‰Si ˆ kŠ ˆ … 1 ‡ k†=‰… 1 †k!Šu1= …1 u†k , where
u ˆ 1 =… 1 ‡ †. The gradient and log-likelihood appear in Greene (2002).
The genesis of the genetic algorithm is in Holland (1975) and has found wide
application to optimization problems. See Goldberg (1989) and Mitchell (1996)
for overviews and guides. Social science applications include Axelrod (1997)
and Andreoni and Miller (1995), among many others.
This index is raised to the third power so as to increase the rate of convergence.
Thus, if strategy A performs 10% better than strategy B, it is 33% more likely to
be chosen as a `parent' than strategy B.
REFERENCES
Alford, Robert D. 1975. Health Care Politics: Ideological and Interest Group Barriers
to Reform. Chicago, IL: University of Chicago Press.
Andreoni, J. and J. Miller. 1995. `Auctions with Arti®cial Adaptive Agents.' Games
and Economic Behavior 10: 39±64.
Austen-Smith, David. 1992. `Strategic Models of Talk in Political Decision Making.'
International Political Science Review 13: 45±58.
Austen-Smith, David and John R. Wright. 1992. `Competitive Lobbying for a Legislator's Vote.' Social Choice and Welfare 9: 229±57.
Austen-Smith, David and John R. Wright. 1994. `Counteractive Lobbying.' American
Journal of Political Science 38: 25±44.
Axelrod, Robert. 1997. The Complexity of Cooperation: Agent-Based Models of
Competition and Collaboration. Princeton, NJ: Princeton University Press.
Bauer, Raymond A., Ithiel De Sola Pool and Lewis Anthony Dexter. 1972. American
Business and Public Policy: The Politics of Foreign Trade, 2nd edn. Chicago, IL:
Aldine Atherton.
Bauman, Patricia. 1976. `The Formulation and Evolution of the Health Maintenance
Organization Policy, 1970±1973.' Social Science and Medicine 10: 129±42.
Bennedsen, Morton and Sven E. Feldmann. 1998. `Persuasion Through the Purse:
How Political Contributions Crowd Out Information.' University of Chicago
Harris School Working Paper Series, pp. 98±101.
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
437
Boorman, Scott A. 1975. `A Combinatorial Optimization Model for Transmission of
Job Information Through Job Contact Networks.' Bell Journal of Economics 6:
216±49.
Browne, William P. 1988. Private Interests, Public Policy, and American Agriculture.
Lawrence, KS: Kansas University Press.
Emerson, Richard and Karen Cook. 1978. `Power, Equity, and Commitment in
Exchange Networks.' American Sociological Review 43: 721.
Goldberg, David E. 1989. Genetic Algorithms in Search, Optimization, and Machine
Learning. Reading, MA: Addison-Wesley.
Golden, Marissa Martino. 1998. `Interest Groups in the Rule-Making Process: Who
Participates? Whose Voices Get Heard?' Journal of Public Administration Research
and Theory 8: 245±70.
Granovetter, Mark. 1977 [1973]. `The Strength of Weak Ties.' In Social Networks: A
Developing Paradigm, ed. Samuel Leinhardt. New York: Academic Press. (Originally published in American Journal of Sociology 78: 1360±80.)
Granovetter, Mark. 1982. `The Strength of Weak Ties: A Network Theory Revisited.'
In Social Structure and Network Analysis, eds Peter Marsden and Nan Lin. Beverly
Hills, CA: Sage.
Greene, William. 2002. Econometric Analysis. Upper Saddle River, NJ: Pearson
Education.
Hansen, John Mark. 1991. Gaining Access: Congress and the Farm Lobby, 1919±1981.
Chicago, IL: University of Chicago Press.
Heclo, Hugh. 1978. `Issue Networks and the Executive Establishment: Government
Growth in an Age of Improvement.' In The New American Political System, ed.
Anthony King. Washington, DC: American Enterprise Institute.
Hojnacki, Marie and David C. Kimball. 1998. `Organized Interests and the Decision
of Whom to Lobby in Congress.' American Political Science Review 92: 775±90.
Holland, John H. 1975. Adaptation in Natural and Arti®cial Systems, 2nd edn. 1992.
Ann Arbor, MI: University of Michigan Press.
Holland, John H. and Miller, John H. 1991. `Arti®cial Adaptive Agents in Economic
Theory.' American Economic Review 81: 365±70.
Holland, Paul and Samuel Leinhardt. 1971. `Transitivity in Structural Models of
Small Groups.' Comparative Group Studies 2: 107±24.
Holland, Paul and Samuel Leinhardt. 1981. `An Exponential Family of Probability
Distributions for Directed Graphs.' Journal of the American Statistical Association
76: 33±65.
Hyman, Joseph. 1971. `Empirical Research on the Demand for Health Services.'
Inquiry 8: 61±71.
Kerwin, Cornelius. 1994. Rulemaking: How Government Agencies Write Law and
Make Policy. Washington, DC: Congressional Quarterly Press.
Knoke, David. 1990. Political Networks: The Structural Perspective. New York:
Cambridge University Press.
Knoke, David, with Franz Urban Pappi, Jeffrey Broadbent and Yutaka Tsujinaka.
1996. Comparing Policy Networks: Labor Politics in the U.S., Germany, and Japan.
New York: Cambridge University Press.
Koenig, Thomas and Thomas Braunninger. 1998. `The Formation of Policy Networks: Preferences, Institutions and Actors' Choice.' Journal of Theoretical Politics
10: 445±72.
438
RATIONALITY AND SOCIETY 15(4)
Kollman, Ken, John H. Miller and Scott E. Page. 1995. `On the Possibility of States as
Policy Laboratories.' Paper presented at the annual meetings of the American Political Science Association, Chicago, Illinois, 31 August to 3 September 1995.
Kollman, Ken, John H. Miller and Scott E. Page. 1998. `Consequences of Nonlinear
Preferences in a Federal System.' Paper presented at the annual meetings of the
Midwest Political Science Association, Chicago, Illinois, April, 1998.
Krackhardt, David. 1992. `The Strength of Strong Ties: The Importance of Philos in
Organizations.' In Networks and Organizations: Structure, Form, and Action, eds
Nitin Nohria and Robert Eccles. Cambridge, MA: Harvard Business School Press.
Laumann, Edward O. and David Knoke. 1987. The Organizational State: Social
Choice in National Policy Domains. Madison, WI: University of Wisconsin Press.
Milbrath, Lester W. 1963. The Washington Lobbyists. Chicago, IL: Rand-McNally.
Mitchell, Melanie. 1996. An Introduction to Genetic Algorithms. Cambridge, MA:
MIT Press.
Padgett, John and Christopher Ansell. 1993. `Robust Action and the Rise of the
Medici, 1400±1434.' American Journal of Sociology 10: 1259±319.
Rasmussen, Eric. 1993. `Lobbying When the Decision-maker Can Acquire Independent Information.' Public Choice 77: 899±913.
Schneider, Mark, Paul Teske, Christine Roch and Melissa Marschall. 1997. `Networks to Nowhere: Segregation and Strati®cation in Networks of Information
about Schools.' American Journal of Political Science 41: 1201±23.
Schwartz, William B. 1972. `Policy Analysis and the Health Care System.' Science 177:
967±9.
Uzzi, Brian. 1996. `The Sources and Consequences of Embeddedness for the Economic Performance of Organizations: The Network Effect.' American Sociological
Review 61: 674±698.
Verbrugge, Lois. 1977. `The Structure of Adult Friendship Choices.' Social Forces 56:
576±97.
Appendix 1: Genetic Algorithm
The genetic algorithm is a computational way of searching a large
solution space that is (roughly) modeled on the evolutionary process.19 The basic idea is that solutions, in the beginning, are randomly generated and then tested. The solutions then `breed', where
probability of breeding is positively related to performance. This
process is iterated, and will usually lead to some convergence on a
particular solution that performs better than other solutions.
More precisely, it is assumed that:
(1) There is some population of problem solutions.
(2) Those solutions are multidimensional and may be represented
numerically on each dimension.
CARPENTER ET AL.: THE STRENGTH OF STRONG TIES
439
(3) After some test or set of tests, new solutions are generated from
the old solutions, where old solutions are spliced together to
generate new solutions. The probability that a given old solution
will be chosen is monotonically related to its performance in the
set of tests. This testing, selection, and splicing process will
usually continue over a period of time.
(4) Often, in addition to the splicing process, there is a mutation
process, where, on a given dimension of a particular `offspring'
strategy, there is some probability that a value on that dimension will be randomly chosen, rather than selected from a parent.
Thus, in the simulation in this article, there is:
(1) A population of N strategies (100 in the simulations summarized).
(2) These strategies are 10 dimensional, one dimension for each
value of . Each dimension may have T= ‡ 1 values (this is
the maximum number of strong ties). Thus, for example, in
the simulations summarized in the article, each dimension may
have a value between 0 and 10. Each NSS may have one of
10…T=† ‡ 1 values (1011 in the simulations summarized).
(3) The strategies are tested by measuring average informedness
over S events (S ˆ 200 in the article). Each event is made up
of R rounds (R ˆ 5 in the simulations summarized). For a
given event, a group's informedness is the fraction of the
rounds in which that group was informed. Thus, if a group
was informed 3 out of the 5 rounds, its informedness in that
event was 0.6. Only the performance of those groups that need
information in a given event and do not have it at the beginning
of the event is counted in assessing the performance of a group.
After S rounds an index of average informedness over those
rounds will be generated. This index is used in choosing the
`parents' of new strategies, where N pairs of parents will be
chosen to generate N new strategies. For each pairing, the probability that a given strategy will be chosen as a parent is proportional to the average informedness of that strategy raised
to the third power.20 For a given pair, an offspring is generated
by randomly choosing ®ve dimensions from each parent. In the
above simulations, this process is repeated 200 times for each
simulation.
(4) There is a mutation probability (= 0.01 in the simulations
summarized) where, for each dimension, a value not from the
440
RATIONALITY AND SOCIETY 15(4)
offspring's parents might be chosen (from a uniform distribution
over the range of possible values of number of strong tie
choices).
DANIEL CARPENTER is Professor of Government at Harvard
University. His book, The Forging of Bureaucratic Autonomy: Reputations, Networks and Policy Innovation in Executive Agencies,
1862±1928, won the 2002 Gladys Kammerer Award and 2002
Charles Levine Award of the American Political Science Association. He is currently engaged in a large-scale theoretical, historical
and statistical analysis of pharmaceutical regulation as carried out
by the U.S. Food and Drug Administration (FDA). He is a
former Robert Wood Johnson Foundation Scholar in Health
Policy at the University of Michigan.
ADDRESS: Littauer Center 224 (North Harvard Yard), Department of Government, Harvard University, Cambridge, MA 02138,
USA.
KEVIN ESTERLING is Assistant Professor of Political Science at
the University of California, Riverside. His research focuses on the
use of policy research and scienti®c information in lobbying politics,
the patterns of information exchange among interest groups, and
the structural conditions for deliberation among health care policy
actors. He is the author of The Political Economy of Expertise: Information and Ef®ciency in American National Politics, which is forthcoming from the University of Michigan Press.
ADDRESS: Department of Political Science, UC ± Riverside, 900
University Ave., Riverside, CA 92521, USA
[email: [email protected]]
DAVID LAZER is Associate Professor of Public Policy and Associate Director of the National Center for Digital Government
Research and Practice at the Kennedy School of Government at
Harvard University. He has extensive research interests in how connectedness affects political and social phenomena, including: small
groups, interest groups, international governance, and presidential
power.
ADDRESS: Kennedy School of Government, 79 John F. Kennedy
Street, Harvard University, Cambridge, MA 02138, USA
[email: [email protected]]